how to write an NRSolve function ?

Questions
, , ,
1

The Mathematica function RSolve has limitations on the form of recurrence formulas that it will handle (these limitatiions seem to give rise to many similar questions on this discussion group).
What existing parts of Mathematica would I need to utilize, to try to write a new function to numerically solve systems of coupled nonlinear recurrence equations?
(call the new function NRSolve)
These nonlinear systems of recurrence equations often arise in macroeconomics.

To be specific, here is a small model (small by macroeconomic standards) that I want to numerically solve using Mathematica, written in a RSolve-type recurrence-equation form:

alpha = 10;
beta = -1/2;
lambda = 1;
epsilon = 3/4;
sigma = 1/2;
At = 200;
Cn = 400;

gL = 2/100;
gMult[t_] := Piecewise[{{1, t <= 1}}, 1 + gL];
mua[t_] := (alpha*Ar[t]^beta - lambda);

eqs = {Cw[t] == (vC[t]*pc[t] + Aw[t - 1]*pa[t] - L[t - 1] + 
      Ar[t]*(epa[t] - pa[t]))/(epc[t] + pc[t]), 
  Cr[t] == (vD[t - 1] + Ar[t - 1]*pa[t])/pc[t], 
  vC[t] == Cw[t] + Cr[t], L[t] == L[t - 1]*gMult[t], vD[t] == L[t], 
  Aw[t] == At - Ar[t], 
  Ar[t] == (Cr[t]*pc[t] + Aw[t - 1]*pa[t] - L[t - 1] - vD[t])/pa[t], 
  pc[t] == pc[t - 1]*gMult[t]*(vC[t]/Cn)^sigma, 
  pa[t] == epa[t]/(1 - mua[t]), 
  epc[t] == epc[t - 1]*gMult[t]*(pc[t]/epc[t - 1])^epsilon, 
  epa[t] == epa[t - 1]*gMult[t]*(pa[t]/epa[t - 1])^epsilon,
  
  Ar[0] == 100, Aw[0] == 100, vC[0] == 400, Cr[0] == 200, 
  Cw[0] == 200, vD[0] == 100, L[0] == 100, pa[0] == 1, pc[0] == 1, 
  epa[0] == 1, epc[0] == 1,
  
  Ar[1] == 100, Aw[1] == 100, vC[1] == 400, Cr[1] == 200, 
  Cw[1] == 200, vD[1] == 100, L[1] == 100, pa[1] == 1, pc[1] == 1, 
  epa[1] == 1, epc[1] == 1}

vars = {Ar[t], Aw[t], vC[t], Cr[t], Cw[t], vD[t], L[t], pa[t], pc[t], 
   epa[t], epc[t]};

(* numerical solution desired for t>=2  *)
2

If you want something akin to NDSolve over a range, RecurrenceTable is a good function to start with. One could create an InterpolatingFunction therefrom, if a continuous rather than discrete result is desired.

Also there may be relevant ideas in the 2008 mathGroup thread at the link below.

3

RecurrenceTable is what you think of as NRSolve.